phyC: Clustering cancer evolutionary trees

Multi-regional sequencing provides new opportunities to investigate genetic heterogeneity within or between common tumors from an evolutionary perspective. Several state-of-the-art methods have been proposed for reconstructing cancer evolutionary trees based on multi-regional sequencing data to develop models of cancer evolution. However, there have been few studies on comparisons of a set of cancer evolutionary trees. We propose a clustering method (phyC) for cancer evolutionary trees, in which sub-groups of the trees are identified based on topology and edge length attributes. For interpretation, we also propose a method for evaluating the sub-clonal diversity of trees in the clusters, which provides insight into the acceleration of sub-clonal expansion. Simulation showed that the proposed method can detect true clusters with sufficient accuracy. Application of the method to actual multi-regional sequencing data of clear cell renal carcinoma and non-small cell lung cancer allowed for the detection of clusters related to cancer type or phenotype. phyC is implemented with R(≥3.2.2) and is available from https://github.com/ymatts/phyC.Author summary: Elucidating the differences between cancer evolutionary patterns among patients is valuable in personalized medicine, since therapeutic response mostly depends on cancer evolution process. Recently, computational methods have been extensively studied to reconstruct a cancer evolutionary pattern within a patient, which is visualized as a so-called “cancer evolutionary tree” constructed from multi-regional sequencing data. However, there have been few studies on comparisons of a set of cancer evolutionary trees to better understand the relationship between a set of cancer evolutionary patterns and patient phenotypes. Given a set of tree objects for multiple patients, we propose an unsupervised learning approach to identify subgroups of patients through clustering the respective cancer evolutionary trees. Using this approach, we effectively identified the patterns of different evolutionary modes in a simulation analysis, and also successfully detected the phenotype-related and cancer type-related subgroups to characterize tree structures within subgroups using actual datasets. We believe that the value and impact of our work will grow as more and more datasets for the cancer evolution of patients become available.